README.md

---
title: Deep Fathom
emoji: 🪨
colorFrom: gray
colorTo: blue
sdk: static
pinned: false
---

# Deep Fathom

Our mission is to make continuous CMMC readiness the norm for the
Defense Industrial Base.

Deep Fathom builds an AI-native compliance platform for CMMC. The
platform takes on the routine work, tracks evidence as teams generate
it, and keeps documentation aligned with the actual state of the
controls. The people behind it are engineers and compliance
practitioners who got tired of treating audits as ceremony.

This Hugging Face org is a placeholder for the open side of that
work. We don't have public artifacts here today. If licensing and
eval results allow, we may publish model adapters in the future, and
datasets later still.

## How we build the models

We choose curation over scale. Roughly 13,000 judge-validated
practitioner pairs do better work than 100,000 unfiltered scrapes
when the domain is this narrow.

Every artifact links back to a reproducible training run. We record
the data version, seed, hyperparameters, and judge configuration so a
later release can be checked against the artifact it claims to
supersede.

When we publish models, we ship LoRA adapters rather than merged
weights. PEFT and vLLM load the base and the adapter together at
runtime. The result is the latest upstream base, our specialty layer
composed on top of it, and the option to swap each layer
independently.

## Principles

1. More rigor. Radically less effort.
2. Frontier AI, applied where it carries weight.
3. Security and compliance treated as one workflow.
4. One truth, shared across the team, the partner, and the assessor.

## Who we work with

Defense contractors and DIB suppliers, including the small businesses
that CMMC was always going to test hardest. MSPs and MSSPs that
deliver compliance services for clients. C3PAOs and assessors who
want shared evidence instead of screenshots.

## Reach

[deepfathom.ai](https://www.deepfathom.ai).